Posts /

gcloud and kubectl on GCP ubuntu images

Twitter Facebook Google+
11 Oct 2019

morning glory!
the well bucket-entangled,
I ask for water
– Fukuda Chiyo-ni

Yet another versioning issue

Versioning is a recurring PITA in my life, it seems. Today I ran (for the second time on the same machine, actually) into an issue with kubectl commands run from my CI server:

Get error executing access token command "/snap/google-cloud-sdk/100/bin/gcloud config config-helper --format=json": err=fork/exec /snap/google-cloud-sdk/100/bin/gcloud: no such file or directory output= stderr=

The actual error message was longer than that, but it basically tells me that my kubectl commands that I run to deploy new versions of some applications in our k8s cluster failed. More specifically, it encountered an error executing acces token command .... What’s that? And how to fix it? Well there are a few ingredients to the mix.


Use the /snap/google-cloud-sdk/current symlink instead of the bespoke snap version number of your gcloud SDK:

$ sed -E -ibak 's/(\/snap\/google-cloud-sdk\/)(.*)(\/bin\/gcloud)/\1current\3/' ~/.kube/config


sed -E             # extended regex
    -ibak          # edit in place, create a ~/.kube/config.bak backup file
    's/(\/snap\/google-cloud-sdk\/)(.*)(\/bin\/gcloud)/\1current\3/' # replace the version number with `current`
    ~/.kube/config # file to update

Authentication in kubectl

If you want more low level details you should definitely go to the official documentation. Here, suffice it so say that I used the following command to configure my kubectl access to the cluster[]:

gcloud container clusters get-credentials <my-cluster-name> --zone <my-cluster-zone>

That’s automagically creating your ~/.kube/config file with everything you need to connect to a specific cluster. Then you’re supposed to kubectl happily ever after, with gcloud being directly invoked to retrieve the authentication tokens automagically. Until..

Oh, snap!

The thing I didn’t expect was a poor synergy with the Google Cloud SDK installation and the produced kubectl configuration. There’s a few ingredients to the mix:

        cmd-args: config config-helper --format=json
        cmd-path: /snap/google-cloud-sdk/100/bin/gcloud

The salvation is provided by the /snap/google-cloud-sdk/current symlink which is always updated by snapd upon installation, making the sed script above fix the issue once and for all. Kinda. There are multiple users on the machine I’m using, and the above configuration is user specific. Hence why I faced this issue twice on the same machine. Welp.

Twitter Facebook Google+